How to allocate pages for a scatter-gather buffer?

How to allocate pages for a scatter-gather buffer?

[Alan Stern]

Jens:

Thanks for your help in tracking down that problem with the usb-storage 
sddr09 driver.

Now a slightly different question.  Suppose I want to allocate some pages
as a scatter-gather buffer for I/O to a device.  I've got the device's
dma_mask; what's the right way to convert that to a GFP bitmask for
alloc_pages() or get_free_pages()?

Also, assuming I'll have to use memcpy() to transfer data into or out of 
the scatter-gather buffer (so it will have to get a kernel virtual mapping 
at some point), am I better off using alloc_pages() followed by 
kmap_atomic()/kunmap_atomic() or get_free_pages()?  Or even kmalloc()?

Alan Stern

Re: How to allocate pages for a scatter-gather buffer?

[Jens Axboe]

On Sun, Nov 09 2003, Alan Stern wrote:
> Jens:
> 
> Thanks for your help in tracking down that problem with the usb-storage 
> sddr09 driver.
> 
> Now a slightly different question.  Suppose I want to allocate some pages
> as a scatter-gather buffer for I/O to a device.  I've got the device's
> dma_mask; what's the right way to convert that to a GFP bitmask for
> alloc_pages() or get_free_pages()?

This is a bit problematic, unfortunately... There really is no good way
to do this, the best you can do is approximate.

> Also, assuming I'll have to use memcpy() to transfer data into or out of 
> the scatter-gather buffer (so it will have to get a kernel virtual mapping 
> at some point), am I better off using alloc_pages() followed by 
> kmap_atomic()/kunmap_atomic() or get_free_pages()?  Or even kmalloc()?

My suggestion would be to do page allocations (not kmalloc()). With kmap
that'll work just fine for you, and it's easy to extend to just use the
right gfp mask if/when we ever get something that works for this.

-- 
Jens Axboe